An Exploration of BigGAN’s Truncation Threshold to Create Artistic (Horrific?) Images of Masks

Some cool images of “masks” generated by a BigGAN model using a high truncation threshold which lets us control the variety of the images. They were generated on this demo page, with a model that was already trained for me. The paper introducing the truncation threshold (or truncation trick) is “Large Scale GAN Training for High Fidelity Natural Image Synthesis” by Andrew Brock, Jeff Donahue, and Karen Simonyan. You can see some weak cases of class leakage (especially some animal features) in these images due to the high truncation threshold. These images are cherrypicked (to look cool) and not a representation of the generator’s real distribution. The truncation threshold used was around 0.6-1.0.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: